ui-first-principles
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an input ingestion surface by accepting user-defined free-text
<scenario>descriptions and external JSON sample files via the--apiargument. This data enters the agent context and is processed through multiple LLM reasoning phases inSKILL.md. To mitigate risks, the architecture employs extensive boundary markers and sanitization viaredact.jsandnormalize-input.jsto replace potentially sensitive data or malicious payloads with opaque tokens before LLM consumption. The capability inventory includes executing dedicated local validation and redaction scripts using standard shell utilities, backed by an automated verification gate (validate-report.js) that enforces compliance and intercepts anomalies.
Audit Metadata